Re: ntp audit spew.

[Paul Moore]

On Tue, Sep 24, 2019 at 9:50 AM Richard Guy Briggs <rgb@xxxxxxxxxx> wrote:
> On 2019-09-23 23:01, Paul Moore wrote:
> > On Mon, Sep 23, 2019 at 5:00 PM Richard Guy Briggs <rgb@xxxxxxxxxx> wrote:
> > > On 2019-09-23 12:14, Paul Moore wrote:
> > > > On Mon, Sep 23, 2019 at 11:50 AM Dave Jones <davej@xxxxxxxxxxxxxxxxx> wrote:
> > > > >
> > > > > I have some hosts that are constantly spewing audit messages like so:
> > > > >
> > > > > [46897.591182] audit: type=1333 audit(1569250288.663:220): op=offset old=2543677901372 new=2980866217213
> > > > > [46897.591184] audit: type=1333 audit(1569250288.663:221): op=freq old=-2443166611284 new=-2436281764244
> > >
> > > Odd.  It appears these two above should have the same serial number and
> > > should be accompanied by a syscall record.  It appears that it has no
> > > context to update to connect the two records.  Is it possible it is not
> > > being called in a task context?  If that were the case though, I'd
> > > expect audit_dummy_context() to return 1...
> >
> > Yeah, I'm a little confused with these messages too.  As you pointed
> > out, the different serial numbers imply that the audit_context is NULL
> > and if the audit_context is NULL I would have expected it to fail the
> > audit_dummy_context() check in audit_ntp_log().  I'm looking at this
> > with tired eyes at the moment, so I'm likely missing something, but I
> > just don't see it right now ...
> >
> > What is even more confusing is that I don't see this issue on my test systems.
> >
> > > Checking audit_enabled should not be necessary but might fix the
> > > problem, but still not explain why we're getting these records.
> >
> > I'd like to understand why this is happening before we start changing the code.
>
> Absolutely.
>
> This looks like a similar issue to the AUDIT_NETFILTER_CFG issue where
> we get a lone record unconnected to a syscall when one of the netfilter
> table initialization (ipv4 filter) is linked into the kernel rather than
> compiled as a module, so it is run in kernel context at boot rather than
> in user context as a module load later.  This is why I ask if it is
> being run by a kernel thread rather than a user task, perhaps using a
> syscall function call internally.

I don't see where in the code that could happen, but I agree that it
looks like it; maybe I'm just missing a code path somewhere.

Is anyone else seeing these records?  Granted my audit test systems
are running chrony, not ntp, but the syscalls/behaviors should be
similar and I can't seem to recreate this.

-- 
paul moore
www.paul-moore.com